Application tourism, AGI, HR Tech and the stack fallacy.

Musings on OpenAI's AI job matching and learning announcement.

OpenAI has announced a plan to move into recruitment and learning. Our little corner of the interwebs is abuzz. I’ve seen several posts that remind me of Dad’s Army’s Private Frazer and Lance Corporal Jones.*

What’s the stack fallacy, you may ask?

Stack fallacy is the mistaken belief that it is trivial to build the layer above yours. The term was coined by Anshu Sharma about a decade ago. It explains why database vendors have been challenged to build applications, and why payroll vendors find it hard to build talent management applications, why apple music sucks, why Microsoft doesn’t quite know what to do with Linkedin and why IBM missed out on the desktop.

(The slide below is from our AI thesis deck). Cartoon from XKCD.

Otter’s obvious corollary: Skipping layers of the stack increases the chance of failure geometrically. This is why printer software is beyond dismal.

What’s application tourism?

This is when a horizontal vendor builds (or announces it plans to build) an application in a space it previously has shown no interest in. After realising it is harder than they thought, they pack up and head on to the next destination, or just go home.

For many years Google was the cruise ship of application tourism.

HR tech is a very popular application tourism destination. HR tech seems easy at first.

The list of horizontal vendors launching HR tech products with fanfair and then later killing them off, or leaving them to wither is long. Facebook built a strong employee comms product, but then lost interest. Google built a some recruitment tech (they hired the best PM I’ve ever worked with, Dmitri Krakovsky, to build it) but then killed it, before it really had a chance. Twitter had a fleeting go too. HR vendors announcing German payrolls (an expensive trip). Both Siebel and Salesforce had a brief dalliance with employee relationship management. Every decade or so Microsoft pops by too.

I’m sure it is an uncanny coincidence, but OpenAI announced their job matching and training endeavours at the same time as Mr and Mrs Trump hosting tech leaders to discuss questions such as AI literacy, inter alia. It’s high season.

My current take on OpenAI

The foundational technology they build is remarkable, but they have significant competition. I sense that OpenAI is casting around for ways to justify its valuation, given that AGI isn’t coming anytime soon to a screen near you. It won’t admit that of course, but when they look in the mirror…

They have been brilliant at raising money, building pioneering tech and setting the industry narrative, but they still seek a business model. Yesterday the talk was that they would be replacing McKinsey and Accenture and the week before that they built a CRM before breakfast, apparently.

The Stack Fallacy means that Fidji Simo has one of the most challenging roles (and exciting jobs) in the software industry.

Some ifs and buts

• If you set your mission to build AGI, it’s hard to get your people fired up about job requisition formats, course proctoring and qualification equivalence. It’s a cultural more than a technical challenge.

• If OpenAI are genuine about their move into building these sort of applications, then it means that they are rowing back on their AGI position. As Doug Monro adroitly puts it “ Why would you need a jobs platform if AGI is doing all the jobs and we are all just writing poetry anyway.”

• If OpenAI are serious about being an application vendor, then other application vendors are likely to seek out alternative foundational technology partners. Why would any recruitment tech vendor work with openAI today if they are really serious about this? Even just announcements like this will create unease. In ecosystems everyone has their place. When platform providers play in the apps space, app vendors don’t hang around, and the platform suffers.

• Employment law is a thing.

• I suspect increasing chatGPT revenues by 1% will have a bigger revenue impact than winning 10% of the job matching space. Putting it another way, even becoming awesome recruitment tech vendor isn’t going to vindicate the company’s valuation.

Back to the announcement

The AI learning play makes a lot more sense than the recruitment play. Most significant tech vendors have some sort of learning academy play. Microsoft, Cisco, IBM, Google, SAP, Salesforce and more. I’m just not convinced OpenAI really wants to be in the skills certification game, again it’s harder than it looks. But it will make Coursera and Udemy a tad nervous.

A press release and a blog post is not a product, but it gets people talking. It may well reignite some of the wrapper v foundation debate I thought we were beyond. It might freeze some spend, and every HR Tech founder is likely to have VCs ask them how will you compete with OpenAI

If they really are serious about this

They need to make some significant acquisitions, and hire folks that understand why this stuff is harder than it looks. Matching people to jobs requires a lot more than generative AI. They have the funding, they have access to great data, deep technical chops and they have a distribution channel right into the laptops of a billion workers. I’ll be watching this with interest.

Final aside

I give thanks every day that LinkedIn is the cinderella of the Microsoft empire. If they actually built decent hr-tech it would be omnipotent.

Speaking of Panic…..

As I usually do, here is a song.

*Dad’s Army was a pioneering British comedy. It still stands up brilliantly.